Connect Claude, Cursor, OpenClaw & Hermes to Honest Invoice: MCP Setup Guide

On an active Pro or Business plan, your AI assistant can create estimates, send invoices, and manage billing from Claude Desktop, Cursor, OpenClaw, Hermes, or another MCP-compatible client. This guide covers a hosted Streamable HTTP connection and a local stdio reference server. Use a dedicated, least-privilege API key created in Honest Invoice Settings; every request acts only on the account associated with that key.

What is MCP?

Model Context Protocol (MCP) is an open standard that lets AI clients such as Claude, Cursor, OpenClaw, and Hermes connect to external services. Honest Invoice supports a hosted Streamable HTTP endpoint for clients that can connect to a remote server, plus a local stdio reference server for clients that launch subprocesses. The available tools are account-scoped, and tool access remains subject to your plan, API-key scopes, rate limits, and your review of any action that sends email or changes records.

Prerequisites

You need a Honest Invoice Pro or Business account and a dedicated API key created in Settings. Node.js 18 or later is required only for the local stdio reference server; the hosted Streamable HTTP option does not require a local checkout or Node.js. Free accounts do not have MCP access. OpenClaw and Hermes are independent third-party products, not affiliated with or endorsed by Honest Invoice, so use their current official documentation and security controls as well.

Step 1: Create a dedicated API key

Open Honest Invoice Settings, create a named API key for this agent, and copy the secret when it is shown. The secret is stored as a hash, scoped to your account, and shown only once. Never put a Supabase service-role key or browser session token in an agent configuration.

Step 2: Configure the hosted Streamable HTTP option

For a client that supports remote MCP servers, add a Streamable HTTP server with the URL https://honestinvoice.com/api/mcp and the header Authorization: Bearer <your-dedicated-key>. Use HTTPS only; never put the key in a URL, browser code, screenshots, or logs. The endpoint is stateless and exposes the same account-scoped tools as the local server. Confirm the client can initialize and list tools before allowing it to create, update, or send anything.

Step 3: Configure Claude Desktop and Cursor

Claude Desktop: open Settings → Developer → MCP Servers and add a remote server using the hosted URL and Authorization header above, or use the local stdio reference server with command npx, args tsx and src/mcp-server.ts, APP_BASE_URL=https://honestinvoice.com, and HONEST_INVOICE_API_KEY in the environment. Cursor: open Settings → Features → MCP, add the same URL-based configuration, or use the local JSON config from the MCP setup page. Restart the client and verify the Honest Invoice tools appear. Store secrets in the client or operating system secret store when available.

Step 4: Configure OpenClaw

In the OpenClaw Control UI, open Settings → MCP → Add server and choose Streamable HTTP. Use https://honestinvoice.com/api/mcp, add Authorization: Bearer <your-dedicated-key>, and keep the server enabled only for the agents or sessions that need it. If you prefer config, add the server under mcp.servers in ~/.openclaw/openclaw.json. Run openclaw mcp doctor honest-invoice --probe after saving, then reload or restart the gateway. OpenClaw is a third-party product; review its tool-policy and approval settings before enabling write or send tools.

Step 5: Configure Hermes Agent

Hermes can connect to the hosted endpoint from ~/.hermes/config.yaml. Add an mcp_servers.honest-invoice entry with url: https://honestinvoice.com/api/mcp, headers.Authorization: Bearer <your-dedicated-key>, and enabled: true. Alternatively, use the local stdio entry with command npx, args [tsx, src/mcp-server.ts], APP_BASE_URL, and HONEST_INVOICE_API_KEY. Start with hermes chat and use /reload-mcp after changes. Keep the key in ~/.hermes/.env or another secret store, and use Hermes tool include/exclude controls to limit exposure.

What your AI agent can do

Once connected on Pro or Business, your AI agent has a full set of account-scoped tools available. create_estimate builds a professional estimate with line items, client details, taxes, and an optional job description for your account. create_invoice does the same for invoices. send_document emails the estimate or invoice from your account. list_documents shows only your account's documents. extract_line_items uses AI to break a plain-English job description into itemized labor and materials. process_lead takes a scraped lead from Craigslist, Nextdoor, or Facebook, extracts line items, builds an estimate for your account, and can email it. scrape_leads is Business-only and automates lead collection.

Real example: responding to a lead

A customer posts on Craigslist looking for an HVAC condenser replacement. You copy their job description and ask your AI agent to create an estimate for a 3-ton Lennox condenser replacement with 6 hours of labor and R-410A refrigerant charge, for client Jane Smith at jane@email.com, with a 7.25% tax rate. Your agent calls create_estimate with the right items pre-populated, then calls send_document to email it. Total time: under 10 seconds. Jane gets a branded, itemized estimate with a payment link before anyone else even opens their invoicing app.

Pro tip: template presets

Save common job descriptions as templates in a text file or prompt library — things like AC replacement 3-ton, Duct cleaning 1500 sq ft, or Electrical panel upgrade 200A. Then just tell your AI agent which template to use and fill in the client name. Your estimate-to-send time drops to about 3 seconds. Use them with the Projects feature in Claude or the .cursorrules file in Cursor for even faster workflows.

The competitive edge

In service businesses, speed kills. The first estimate a prospect receives wins the job about 60% of the time, based on industry sales data from home services, HVAC, and electrical contractors. Most contractors take 20 to 60 minutes to write and send an estimate — by then, someone else already has their foot in the door. With MCP connected, your AI agent can respond to a lead in seconds. Not minutes. Not hours. Seconds. That is how you turn a Craigslist post into a signed job while your competition is still opening QuickBooks.

Lead scraping with the webhook API

The process_lead tool and its REST endpoint at /api/mcp/leads/webhook handle the entire pipeline: receive a scraped lead, extract line items with AI, create an estimate, record the response, and email it to the lead. Connect an external scraping service or cron job to POST leads in real-time. On the Business plan, this runs fully automated — your AI agent responds to every lead the moment it appears, with zero human latency. That is how you win jobs before your competition even sees them.

Security, legal, and troubleshooting notes

Use HTTPS and a dedicated key with only the read, write, send, ai, or leads scopes you actually need. Rotate keys periodically and revoke them immediately if a workstation, gateway, workspace, or log may have exposed one. Review tool calls before sending documents or contacting leads. You are responsible for accurate invoice data, tax treatment, customer consent, privacy obligations, email and anti-spam compliance, and the terms of any source platform used for lead collection. Honest Invoice does not provide legal, tax, or compliance advice, and OpenClaw, Hermes, Claude, and Cursor are independent third-party products. If a connection fails, verify the endpoint URL, HTTPS certificate, Authorization header, plan status, key scope, and the client’s transport selection. For local stdio, also verify Node.js and the working directory. Never paste a real secret into support tickets or documentation.